Weights & Biases

The Weights & Biases integration connects Althea’s code execution workflows to your W&B account. Once connected, code agents and remote training jobs can use your WANDB_API_KEY to log metrics, configurations, checkpoints, and experiment artifacts.

Setup

Open settings by clicking your profile icon in the top right corner.

Navigate to Weights & Biases under Integrations.

Click Get API Key on Wandb to open the W&B authorization page. Copy your API key from W&B, paste it into Althea, and click Connect.

Althea verifies the key against the W&B API before storing it. When the connection succeeds, the integration shows the connected W&B username.

How It Works

Althea stores the API key as a user secret. When a code workflow needs experiment tracking, the runtime resolves that secret and exposes it to the job as:

WANDB_API_KEY

The agent can then initialize W&B inside generated training scripts:

import os
import wandb

if os.environ.get("WANDB_API_KEY"):
    wandb.init(project="my-experiment", config={"learning_rate": 0.001})
    wandb.log({"loss": 0.42})

For remote-server jobs, the submission helper also forwards the key into the remote job environment when it is available.

Usage

Ask Althea to use W&B when running experiments:

“Train a small ResNet on CIFAR-10 and log the training metrics to Weights & Biases.”

“Run three learning-rate sweeps and create a W&B run for each configuration.”

“Use the remote server, save checkpoints, and upload the final model artifact to W&B.”

If your GitHub integration is also connected, Althea can combine both: push experiment code to a private repository and log runs to W&B for monitoring and comparison.

Disconnect

Return to the Weights & Biases integration panel and click Disconnect.

Disconnecting revokes the stored W&B secret from Althea. Future code-agent and remote-server jobs will no longer receive WANDB_API_KEY.